Gene structure, regulatory control, and evolution of black widow venom latrotoxins

Abstract

Black widow venom contains α-latrotoxin, infamous for causing intense pain. Combining 33 kb of Latrodectus hesperus genomic DNA with RNA-Seq, we characterized the α-latrotoxin gene and discovered a paralog, 4.5 kb downstream. Both paralogs exhibit venom gland specific transcription, and may be regulated post-transcriptionally via musashi-like proteins. A 4 kb intron interrupts the α-latrotoxin coding sequence, while a 10 kb intron in the 3′ UTR of the paralog may cause nonsense-mediated decay. Phylogenetic analysis confirms these divergent latrotoxins diversified through recent tandem gene duplications. Thus, latrotoxin genes have more complex structures, regulatory controls, and sequence diversity than previously proposed.

1. Introduction

Venom proteins can evolve rapidly through gene duplication and adaptation in response to selection imposed by diverse and co-evolving prey [1–4]. Venom toxins also have important biomedical applications, including receptor characterization and drug discovery [5,6]. However, few studies have examined the structure and regulation of genes encoding venom proteins [7]. For spiders, one of the largest venomous clades, nearly all sequences come from venom gland cDNAs [8,9]. Thus, the roles of gene duplication, alternative splicing, and regulatory controls in generating venom molecular complexity are poorly understood.

The spider genus Latrodectus includes black widows (multiple species) and the Australian red-back spider (L. hasselti), the venoms of which have potent neurotoxic effects on vertebrates. Latrodectus venom contains latrotoxins, a family of neurotoxic proteins that share a unique N-terminal domain flanked by 11–20 ankyrin motif repeats [6,8,10]. Four latrotoxins have been functionally characterized from L. tredecimguttatus, and while three (α-latroinsectotoxin, α-latrocrustotoxin and δ-latroinsectotoxin) elicit neurotransmitter release in arthropods [11], α-latrotoxin forms calcium channels in vertebrate pre-synaptic neuronal membranes, thereby triggering massive neurotransmitter exocytosis [12–14]. α-Latrotoxin is responsible for the extreme pain resulting from black widow bites [6], and is important for studying neurosecretion, and hence has received considerable scientific attention.

While Latrodectus venom contains multiple latrotoxins, they have only been identified in three genera of the family Theridiidae, suggesting latrotoxins represent a recently expanded protein family [8,15,16]. Evidence from RNA-Seq data indicates ≥20 divergent latrotoxins are expressed in venom glands of the Western black widow (Latrodectus hesperus) and most are phylogenetically distinct from the functionally characterized latrotoxins [16]. We hypothesize these transcripts are encoded by distinct loci that are spatially clustered in the genome and that their transcription and translation is tightly controlled to ensure strong venom gland-specific expression.

We sequenced 33 kb from the L. hesperus genome encompassing the α-latrotoxin gene, which we integrated with venom gland expression data (RNA-Seq and Expressed Sequence Tags (ESTs)) to reveal α-latrotoxin’s gene structure, quantify its expression, and investigate its regulation. This revealed a divergent, highly expressed latrotoxin paralog 4.5 kb downstream of α-latrotoxin, and long introns and putative regulatory elements in both paralogs, providing novel insights into latrotoxin evolution and production.

2. Materials and Methods

2.1. Genomic Sequencing

A genomic library was constructed from eight L. hesperus females to cover the estimated 1261 Mb genome [17]. Library construction details are in Supplementary Material and Ayoub et al. [18]. α-Latrotoxin primers were used to PCR-screen the genomic library, revealing three positive clones. The clone with the smallest insert (estimated from a BamHI digest) was used to make a shotgun library from three complete, separate digests (EcoRI, PstI and EcoRV). Resulting fragments were ligated into pZErO^™-2 plasmids (Invitrogen) and electroporated into TOP10 E. coli. For each digest, a library of 192 clones was screened for insert size. 1–2 kb inserts were sequenced using Sp6 and T7 primers. Sequences were edited and assembled in SEQUENCHER 4.2 (Gene Codes Corp.). Primer walking was employed to complete insert sequencing. The complete sequence was deposited at NCBI (Accession KM382064).

2.2. Genomic Annotation

Open Reading Frames (ORFs) were predicted from the assembled genomic insert sequence using getORF, retaining ORFs encoding ≥30 amino acids. Predicted proteins were subjected to BLASTp searches against NCBI’s non-redundant (nr) protein database. Gene structure and expression were explored using female L. hesperus RNA-Seq data [19] from three tissues: (1) venom gland, (2) total silk gland tissues, and (3) cephalothorax minus venom glands [16,19]. This included 133 million high quality 75–100 bp paired-end sequence reads collectively. Trinity [20] was used to assemble tissue-specific reads, and overlapping sequences across libraries were merged using CAP3 [19]. Tophat 2.0.8b [21] was used to align RNA-Seq reads from tissues to the genomic sequence to map introns, which were visualized with Integrative Genomics Viewer (IGV) 2.3 [22]. Cufflinks 2.0.2 [23] was used to assemble transcripts from Tophat mappings, filtering out isoforms representing <10% of major isoform abundance. Cuffmerge was used to merge assemblies from tissues, and Cuffdiff was used to produce expression estimates for transcripts in each tissue. OrfPredictor [24] was used to translate proteins from transcripts in the frame of the top nr BLASTx hit. We also mapped L. hesperus venom gland ESTs [25] (NCBI dbEST Accessions JZ577614 - JZ578096) to the genomic sequence.

The insert sequence was annotated with MAKER [26]. Gene prediction was performed using Augustus under Drosophila melanogaster training parameters and mapping: (1) de novo Trinity assembled transcripts [19] (2) Cufflinks transcripts predicted from the insert, (3) venom gland ESTs [25] and (4) latrotoxins translated from Trinity transcripts. RepeatMasker [27] was used to predict repetitive elements and low complexity regions with the D. melanogaster database. The programs NNPP 2.2, Match 1.0, UTRscan, and PipMaker [28–31] were used to predict regulatory elements (Supplementary Materials).

2.3. Evolutionary Analyses

We obtained latrotoxin proteins from NCBI’s nr and extracted L. tredecimguttatus latrotoxins [8] from the transcriptome shotgun assembly archive using tBLASTn (e-value cutoff 1e-20), with the first 320 amino acids from L. hesperus latrotoxins [16] as queries. L. tredecimguttatus translations were obtained with OrfPredictor [24]. We aligned these sequences with L. hesperus latrotoxins predicted by Cufflinks from the genomic insert and de novo assembled from RNA-Seq data [16,19] with COBALT [32]. Phylogenetic analyses were performed with Mr. Bayes 3.2.2 [33], with a “mixed” amino acid model and gamma distribution, running 5 × 10⁶ generations. Clade posterior probabilities were from a 50% majority-rule tree, excluding the first 25% of trees. We produced a nucleotide alignment with PAL2NAL [34] from the protein alignment (Supplementary Materials). PAML 4 [35] was used to test for positive selection among sites by comparing Model M7 (several dN/dS categories between 0 and 1), and model M8 (adds category with unconstrained dN/dS), and among sites along specific branches by comparing models A and A_ω2=1. MEME [36] was also employed to find sites under diversifying selection across the latrotoxin phylogeny with an FDR < 0.05.

3. Results

3. 1. Venom Expressed Tandem Latrotoxins

We sequenced 33,342 bp of L. hesperus DNA from a genomic library clone containing the α-latrotoxin gene. Sanger sequencing and shotgun assembly of 236 subclones resulted in six contigs 968–8347 bp long, and the assembly was finished through primer walking. getORF predicted 129 proteins in both directions, 15 with significant BLASTp hits (Table 1). Two proteins (1401 and 1393 amino acids) had top BLASTp hits to α-latrotoxin. The upstream predicted protein had a top hit to L. hesperus α-latrotoxin with 99% identity and represents the α-latrotoxin locus, while the downstream latrotoxin had a best match to Steatoda grossa α-latrotoxin at 44% identity, and represents an adjacent paralog (Table 1). The translated tandem L. hesperus latrotoxins share 43% protein sequence identity.

Table 1.

BLASTp hits for the proteins predicted by the getORF program from the complete L. hesperus 33 kb 119_P5 genomic insert sequence. Start and end columns indicate position of the ORF in the sequence. The final column indicates whether the best-hit NCBI accession is a transposable element.

Accession	Description	e-value	start	end	length	TE?
AGD80166.1	Alpha-latrotoxin (Latrodectus hesperus)	0.0	2271	6473	4203	N
EEZ99596.1	hypothetical protein TcasGA2_TC002110 [Tribolium castaneum]	7.00e–30	7596	7159	436	Y
EEZ99596.1	hypothetical protein TcasGA2_TC002110 [Tribolium castaneum]	3.00e–17	7907	7542	364	Y
XP_001602535.2	hypothetical protein LOC100118603 [Nasonia vitripennis]	1.00e–33	15195	14575	619	Y
AGD80173.1	Alpha-latrotoxin (Steatoda grossa)	0.0	18794	22972	4179	N
EJY57592.1	AAEL017053-PA [Aedes aegypti]	8.00e–11	23965	23798	166	N
AAN87269.1	ORF [Drosophila melanogaster]	4.00e–08	24578	24465	112	Y
EFA02763.1	hypothetical protein TcasGA2_TC008496 [Tribolium castaneum]	5.00e–26	24917	24633	283	Y
WP_006706426.1	Pao retrotransposon peptidase protein, partial [Candidatus Regiella insecticola]	4.00e–39	25890	25456	433	Y
XP_003245995.1	hypothetical protein LOC100570266 [Acyrthosiphon pisum]	2.00e–31	26481	26002	478	Y
EFA13346.1	hypothetical protein TcasGA2_TC002325 [Tribolium castaneum]	1.00e–32	26956	26504	451	Y
XP_003699175.1	uncharacterized protein LOC100874762 [Megachile rotundata]	1.00e–10	27862	27305	556	Y
XP_002012451.1	GI14275 [Drosophila mojavensis]	8.00e–10	28257	27931	325	N
CAA35587.1	unnamed protein product [Drosophila melanogaster]	8.00e–59	30831	29839	991	Y
AAX07244.1	reverse transcriptase [Liobuthus kessleri]	3.00e–23	31174	30812	361	Y

Mapping of L. hesperus RNA-Seq data [19] with Tophat to the insert indicated that both latrotoxin genes are highly transcribed in venom glands (Figure 1), with Cufflinks venom gland FPKM (fragments per kilobase per million library reads) of 80157 for α-latrotoxin and 81563 for the downstream paralog. FPKM in cephalothorax and silk glands was far lower for both latrotoxins (α-latrotoxin: cephalothorax=1318, silk glands=1442; downstream paralog: cephalothorax=76, silk glands=0).

Figure 1.

Mapping of RNA-Seq data to the 33 kb L. hesperus genomic reference sequence showing strong venom gland specific expression of tandem latrotoxin paralogs. Top ruler shows location in the reference sequence. Representative reads from the L. hesperus venom gland (top panel), cephalothorax (middle) and silk gland (bottom) RNA-seq libraries are shown mapped to the reference using Tophat with visualization in IGV. Reads mapping across intron boundaries are connected by thin lines and show location of splice junctions. At the bottom the gene structure of the two latrotoxins encoded by the fosmid sequence as determined by Cufflinks is diagrammed. In exonic regions, coding sequences are in gray, UTRs in black. The dashed line indicates the extended 3′ UTR region predicted by the Trinity assembled transcript. The position of the mariner element is indicated in the α-latrotoxin intron, and the RepeatMasker predicted retrotransposon positions are indicated in red in the intron of the downstream paralog.

Venom gland RNA-Seq and EST evidence revealed one 4241 bp phase 1 intron in α-latrotoxin’s 25^th codon from the stop codon and a 10,005 bp intron in the 3′ UTR of the adjacent latrotoxin (Figures 1–2, Table 2). The Cufflinks α-latrotoxin transcript had a 4257 bp coding region, a 316 bp 5′ UTR and a 2805 bp 3′UTR (Table 2). The Trinity α-latrotoxin transcripts consisted of two overlapping (by 25 bp) fragments, one with a 150 bp 5′UTR and the other with a 4828 bp 3′UTR. The downstream paralog’s Cufflinks transcript contained a 4182 bp coding region, which was collinear with the genomic sequence, indicating no intron interrupting the coding region. The Cufflinks transcript for this locus had a 675 bp 5′ UTR and a 341 bp 3′ UTR flanking the 10,005 bp intron located 57 bp into the 3′UTR. The Trinity transcripts comprised two overlapping (by 25 bp) fragments, which lacked the first 10 bp of coding sequence, and with the downstream transcript 6 bp shorter than the Cufflinks transcript, but agreeing in intron position. A single bp insertion in the Trinity transcript produced a frameshift and premature truncation, but mapped reads suggested its incorrect assembly. The end of α-latrotoxin’s 3′ UTR and the putative transcription start site of the adjacent latrotoxin were separated by 4545 or 2522 bp, depending if the Cufflinks or Trinity α-latrotoxin was considered. The MAKER annotation also supported two tandem latrotoxin paralogs, but Augustus gene predictions suggested multiple introns in both paralogs, disagreeing with the transcript evidence (Figure 2).

Figure 2.

MAKER-annotated L. hesperus genomic sequence illustrating latrotoxin introns containing transposable elements. The top annotation track shows Augustus predicted gene structure. The middle track shows alignments of RNA-Seq assembled transcripts and ESTs to the fosmid (Cufflinks assembled transcripts labeled with prefix “CUFF”, Trinity assembled transcripts labeled with prefix “venom_comp”, all other sequences are L. hesperus venom gland ESTs labeled with their GenBank Accession numbers), while the bottom section shows repeat regions identified by RepeatMasker (all features annotated in this track from 0–19 kb are simple repeat/low complexity regions; features between 22 kb–32 kb are retrotransposon fragments). The position of the putative Tc1/mariner family transposon not predicted by RepeatMasker is shown in green.

Table 2.

Locations of latrotoxin gene features in the L. hesperus genomic insert sequence. In cases where the predicted length of a feature varies when measured using the Trinity predicted transcript or the Cufflinks predicted transcript, both values are shown. The shaded portion gives values for the upstream α-latrotoxin locus, the unshaded for the downstream latrotoxin paralog.

Feature	Start	End	Length
5′ UTR exon (Cufflinks)	1955	2270	316
5′ UTR exon (Trinity)	2121	2270	150
Coding exon	2271	6450	4180
Intron	6451	10691	4241
Coding exon	10692	10768	77
3′ UTR exon (Cufflinks)	10769	13573	2805
3′ UTR exon (Trinity)	10769	15596	4828
5′ UTR exon	18119	18793	675
Coding exon	18794	22975	4182
3′ UTR exon	22976	23032	57
Intron	23033	33037	10005
3′ UTR exon (Cufflinks)	33038	33321	284
3′ UTR exon (Trinity)	33038	33315	278

3.2. Latrotoxin Transposable Elements

RepeatMasker identified seven transposable elements (TEs) from 68 to 1215 bp in the 3′ UTR intron of the downstream latrotoxin. Five belong to the Pao retrotransposon family and two were long interspersed nuclear elements (LINEs) in the Jockey retrotransposon family (Figure 2, Table 3). Most getORF proteins had significant BLAST hits to TEs (Tables 1, 3). These elements, or closely related copies, showed high expression in silk and cephalothorax tissues, but not in venom glands (Supplementary Materials, Table 2).

Table 3.

Repeats found in the L. hesperus genomic insert sequence. For simple repeats the repeat motif is shown. All repeats were identified by RepeatMasker with the exception of the putative Tc1/mariner superfamily DNA transposon (shaded row) identified as discussed in the text.

Repeat Class/Family	Position in insert sequence		Matching repeat
	Begin	End
Simple Repeat/ Low complexity	765	806	(ATTTA)n
	2102	2143	(A)n
	8662	8691	(AG)n
	9476	9532	(ATTTT)n
	11119	11213	(TATTAAT)n
	13027	13076	A-rich
	15841	15890	(TTA)n
	16588	16656	A-rich
	17928	17978	(TAATTTT)n
LTR/Pao/BEL	23817	25031
	25415	26100
	26331	26839
	27242	27324
	28119	28193
LINE/Jockey	30008	30075
	30515	31256
Tc1/mariner	7011	8435

A putative TE not detected by RepeatMasker was located within the α-latrotoxin intron. BLASTn of this intron found significant hits to L. hesperus genomic sequences surrounding the dragline silk MaSp1 and MaSp2 encoding genes. The top BLASTn hit was to two different regions (1304 or 1429 bp at 92% identity) of a MaSp2 containing clone (EF595245) from the library used in this study [18]. A 652 bp BLASTn match was found to a different L. hesperus genomic region from a MaSp1 containing clone (EF595246). A Cufflinks transcript for this TE in α-latrotoxin’s intron (positions 7014–8433), or closely related copies, had much higher expression in silk glands and cephalothorax than in venom glands (FPKM: venom gland=1447, cephalothorax=304868, silk gland=838794). BLASTx of this transcript to NCBI revealed homology to transposases of the Tc1/mariner DNA transposon superfamily [37], in agreement with getORF predictions (Table 1). EMBOSS einverted found 41 bp inverted repeats flanking coordinates 7011–8435, indicating the mariner-like transposon in the intron in α-latrotoxin is 1425 bp.

3.3. Latrotoxin Regulatory Elements

Twenty-nine promoters were predicted in the 33 kb insert (Supplementary Materials, Table 3), with a score ≥ 0.95. None were within 600 bp of a TSS (defined by transcript termini). When run on 5′ UTRs, 3′ UTRs, or putative promoters upstream of TSS, Pipmaker found no conserved regions between latrotoxin paralogs. Several TF binding sites were predicted by Match 1.0 in the introns of both paralogs, but their orientation was opposite of latrotoxin genes, and functions of the TFs binding these sites had no obvious connection to venom production (Table 4).

Table 4.

Transcriptional regulatory protein binding sites identified in latrotoxin introns using Match 1.0. Shown are position within intron sequence, orientation, name and functional information.

Position	Strand	Name	Function
563	-	CF2-II	Late activator in follicle cells during chorion formation
2463	-	BR-C Z1	Ecdysone response during metamorphosis
2842	-	BR-C Z4	Ecdysone response during metamorphosis
3930	-	BR-C Z4	Ecdysone response during metamorphosis
895	-	Elf-1	Epidermal differentiation
4447	-	Croc	Head development
4664	-	BR-C Z1	Ecdysone response during metamorphosis

UTRscan predicted upstream open reading frames (uORFs) in the 5′ UTRs of both latrotoxin transcripts, an internal ribsosome entry site (IRES) in the 5′ UTR of α-latrotoxin, and musashi binding elements (MBEs) in the 5′ UTR of the downstream latrotoxin (Supplementary Materials, Table 4). The 3′ UTR from the Cufflinks α-latrotoxin contained multiple MBEs, with additional MBEs and a Brd-Box motif in the longer 3′ UTR from the Trinity transcript. Both α-latrotoxin transcript 3′ UTRs terminated in a predicted polyadenylation signal, suggesting possible alternative transcripts with different 3′ UTRs. The Cufflinks 3′ UTR of the downstream latrotoxin gene had one predicted terminal polyadenylation site.

3.4. Latrotoxin Family Evolution

Most latrotoxins from L. hesperus group in highly supported clades with orthologs from L. tredecimguttatus, with five exceptions, including the paralog adjacent to α-latrotoxin (Figure 3). The adjacent latrotoxin genes are closely related, but not sister paralogs, with α-latrotoxin’s downstream paralog being more closely related to another paralog. The Cufflinks-translated L. hesperus α-latrotoxin falls within a clade of orthologs. A test for sites evolving under positive selection using all latrotoxins (M7 vs. M8: 2 Δ lnL=9.113; df=2; p=0.0104) was significant, with an additional M8 category of sites with an average dN/dS of 1.618. However, no sites had a posterior probability > 0.592 in the site class under positive selection. A branch-site test was applied specifically to two branches subsequent to the duplication event separating α-latrotoxin from the downstream paralog (Figure 3), and a model allowing for positive selection on these branches was a better fit to the data (MA vs. MA_ω2=1: 2 Δ lnL=14.219; df=1; p= 0.0001), with two codons (389 and 886) having a posterior probability > 0.95 of dN/dS > 1. MEME indicated 9 codons under positive selection, with most clustered towards the C-terminus of the latrotoxin protein (Supplementary Material).

Figure 3.

Evolutionary relationships of latrotoxin paralogs indicating the relatively recent duplication of tandem latrotoxins. Shown is a midpoint rooted Bayesian phylogenetic tree of latrotoxin proteins. Numbers at nodes indicate clade posterior probabilities (PP). The node marked by an asterisk has a PP of 1.00. Sequences from L. tredecimguttatus are in red and L. hesperus in blue; sequences from L. geometricus and S. grossa are in orange or green, respectively. Arrows indicate the tandem sequences assembled by Cufflinks from this study. All other L. hesperus and L. tredecimguttatus sequences were from de novo assembly of RNA-Seq reads. Shaded beige boxes label clades that contain a functionally characterized paralog. The larger gray shaded clade contains the two genomic insert latrotoxins and more closely related sequences, including all α-latrotoxins. Branches tested for sites under positive selection are indicated by the number 1 in purple text.

4. Discussion

Characterization of 33 kb of the Western black widow (L. hesperus) genome revealed the complete α-latrotoxin gene and a divergent closely related paralogous locus, and both are highly expressed in venom glands at similar levels. These loci represent two of ≥ 20 potential L. hesperus latrotoxin genes with venom-gland specific expression [16]. These adjacent paralogs occur 4.5 kb apart, providing evidence of a tandem gene duplication. Both latrotoxin loci encode large proteins, but contrary to previous findings in L. tredecimguttatus [38], introns are present in their coding sequence or 3′ UTR. The latrotoxin coding and 3′ UTR introns are approximately 5x and 7.5x the average length of introns from corresponding regions in Drosophila melanogaster [39]. However, these introns are similar to average lengths recently reported from spider genomes [40]. While energetically costly to transcribe [41], introns can increase gene expression when they contain regulatory enhancers, or by direct interaction of the transcriptional and spliceosomal machinery [42].

Latrotoxin UTRs have several predicted regulatory sequences that may function in post-transcriptional regulation. The musashi mRNA binding protein exerts translational control and modulates cell division and cell cycle progression [43], and the multiple UTR musashi binding elements predicted in both latrotoxins may regulate their expression. Haney et al. [16] reported a protein with a significant BLAST hit to a putative RNA-binding musashi protein among 695 L. hesperus venom gland specifically expressed transcripts, with venom gland biased expected read counts per million reads among tissues (cephalothorax=54.64; silk=5.9; venom=2647.02), supporting the role of MBEs in latrotoxin regulation. Upstream ORFs were also predicted in the 5′ UTRs of both paralogs, which may modulate expression through effects on translation efficiency [44]. Furthermore, the presence of a 3′UTR intron may target the downstream latrotoxin’s transcript to the nonsense mediated decay (NMD) pathway, greatly reducing translation [45]. In mammals, this typically occurs when the 3′ most exon-exon junction is >50–55 bp downstream of the stop codon [46], and the splice junction in the 3′ UTR of the downstream latrotoxin is 57 bp beyond the stop codon. Despite both genes being expressed at similar levels in venom glands, the downstream latrotoxin paralog was not identified in venom by mass spectrometry [16], consistent with its transcripts being targeted to the NMD pathway. While NMD has not been studied in arachnids, BLASTp of three components of this pathway (UPF1-3) from Homo sapiens to the L. hesperus transcriptome identified expressed homologs in spiders (e-values of best hits: UPF1 = 0.0; UPF2 = 0.0; UPF3 = 7e-58).

The latrotoxin paralogs have been subject to several transposable element (TE) insertions, mostly LTR retrotransposons belonging to the BEL/Pao family, and nonLTR retrotransposons of the Jockey family, both widespread in eukaryotic genomes and occurring within the downstream paralog’s 3′ UTR intron [47,48]. The identified TEs also exhibit tissue-specific expression, with much lower expression in the venom gland, relative to silk and cephalothorax tissues. A large TE was found in α-latrotoxin’s intron, with homology to the Tc1/mariner DNA transposon superfamily, which is amenable to interspecific horizontal transfer [37]. This transposon may have recently proliferated in the L. hesperus genome, as closely related sequences (up to 92% identity over 1400 bp) were identified in two other sequenced clones from this same genomic library, representing different loci [18]. Moreover, the mariner-like TE composes 33% of the length of α-latrotoxin’s intron, and its recent insertion may be linked to this intron’s origin [49,50]. Although a complete transposase ORF was not present in this TE, BLASTp of this sequence to the L. hesperus transcriptome reveals expressed transcripts encoding complete transposase, suggesting these mariner-like elements are actively transposing. While the intronless nature of other latrotoxins [38,51] is consistent with a recent origin of introns in L. hesperus latrotoxins, the methods used to identify introns in L. tredecimguttatus latrotoxins (PCR-RFLP [38]) may have missed small introns, or those outside coding regions.

Nearly all latrotoxin transcripts assembled from RNA-Seq data in L. hesperus had a closely related ortholog in L. tredecimguttatus (also assembled from RNA-Seq data), implying the latrotoxin gene family had diversified through numerous gene duplications in the most recent common ancestor of these two species or in an earlier ancestor. Yet the paralog downstream of α-latrotoxin did not appear to have an L. tredecimguttatus ortholog, suggesting it is either not evolutionarily conserved, is a recent duplicate for which the closely related paralog in L. hesperus was not sampled, or lacks or exhibits variable expression in L. tredecimguttatus. The adjacent latrotoxins are not sister paralogs, but occur in the same sub-clade, supporting recent duplication events via non-homologous crossing-over, and we expect that many other latrotoxins are clustered in the Latrodectus genome. While latrotoxins largely appear to be evolving under purifying selection, our analyses found limited evidence of positive selection on a few codons, suggesting the possible adaptive evolution of latrotoxins. Clearly, functional assays are needed to test the toxin activity of these recently described paralogs, which is currently unknown. The high level of sequence divergence between α-latrotoxin and its mostly closely related paralogs would suggest functional variability. While gene duplication and selection are implicated as important processes in venom evolution, as has been shown in spiders and other venomous taxa [52–55], functional data will add to our understanding of how these processes impact venom phenotypes.

Highlights.

• α-latrotoxin is a black widow venom toxin causing massive neurotransmitter release

• Genomic and RNA-Seq data reveal α-latrotoxin’s gene and a tandem divergent paralog

• Tandem latrotoxin paralogs are highly and specifically transcribed in venom glands

• Latrotoxins have 4–10 kb introns containing multiple transposable elements

• Latrotoxins are divergent in sequence and may be post-transcriptionally regulated

Abbreviations

FPKM

fragments per kilobase per million library reads

TE

transposable element

LINE

long interspersed nuclear element

MaSp

major ampullate silk protein

TSS

transcription start site

TF

transcription factor

MBE

musashi binding element

LTR

long terminal repeat

NMD

nonsense mediated decay

FDR

false discovery rate